r.mfilter should _respect_ i.fft outputs for use with i.ifft

[nikos]

# fft of some raster image
i.fft in=a real=a_real imaginary=a_imaginary

# do something with the real_fft_image, e.g. r.mfilter
r.mfilter in=a_real filter=some_filter out=a_real_filtered

# perform i.ifft
i.ifft real=a_real_filtered imaginary=a_imaginary output=a_via_fft_filtered

ERROR: Unable to open real-image in the 'cell_misc' directory. Raster map probably wasn't created by i.fft

# questions

Is it fair enough to ask for this support?
Or is there a reason why _r.mfilter_ alters the i.fft-products in such a way that _i.ifft_ can't handle?

Nikos

[nikos]

Well, I bombed (again) in an issue that I had seen in the past but did not remember of it anymore.

Read ​Question about MODIS band 5 destriping?

[glynn]

Replying to nikos:

Is it fair enough to ask for this support?

It's already present in 7.0, but the interface had to be changed (separate maps for the real and imaginary components), so it's unlikely to be back-ported.

Or is there a reason why _r.mfilter_ alters the i.fft-products in such a way that _i.ifft_ can't handle?

i.fft stores the actual frequency-domain data in auxiliary data files. The output map only exists for viewing with d.rast. Apart from anything else, a raster map can only store real values, and the FFT results are complex values.

i.ifft operates upon the auxiliary data files, not the map itself, so it only works with maps which were generated by i.fft. The only processing which can be performed on the frequency-domain data is to apply a mask (i.ifft explicitly applies the MASK raster if it exists).

[nikos]

Should this ticket be marked as will-not-be-fixed for grass64?

Nikos